Users of computing devices (e.g., laptops, cellular phones, and personal digital assistants) often need to communicate in real time. A common form of real-time communications is provided by instant messaging services. An instant messaging service allows participants at endpoints to send messages and have them received within a second or two by the other participants in a conversation. The receiving participants can then send responsive messages to the other participants in a similar manner. To be effective, a real-time conversation relies on the participants' becoming aware of, reviewing, and responding to received messages very quickly. This quick response is in contrast to conventional electronic mail systems in which the recipients of electronic mail messages respond to messages at their convenience.
To support real-time communications, communications applications typically need to establish and manage connections (also referred to as sessions or dialogs) between computing devices. A session is a set of interactions between computing devices that occurs over a period of time. As an example, real-time communications applications such as MESSENGER or VoIP establish sessions between communicating devices on behalf of users. These applications may use various mechanisms to establish sessions, such as a “Session Initiation Protocol” (“SIP”). SIP is an application-level control protocol that computing devices can use to discover one another and to establish, modify, and terminate sessions between computing devices. SIP is a proposed Internet standard. The SIP specification, “RFC 3261,” is available at via the Internet Engineering Task Force (“IETF”) web site.
A SIP network comprises entities that can participate in a dialog as a client, server, or both. SIP supports four types of entities: user agent, proxy server, redirect server, and registrar. User agents initiate and terminate dialogs by exchanging messages with other SIP entities. A user agent can be a user agent client (“UAC”), which is a device that initiates SIP requests, or a user agent server (“UAS”), which is a device that receives SIP requests and responds to such, requests. As examples, “IP-telephones,” personal digital assistants, and any other type of computing device may be user agents. A device can be a UAC in one dialog and a UAS in another, or may change roles during the dialog. A proxy server is a device that acts as a server to clients and a client to servers. In so doing, proxy servers intercept, interpret, or forward messages between UACs and UASs. A redirect server is a device that accepts a SIP request and generates a response directing the UAC that sent the request to contact an alternate network resource. A registrar is a server that accepts registration information from user agents and informs a location service of the received registration information.
SIP supports two message types: requests, which are sent from a UAC to a UAS, and responses, which are sent from a UAS to a UAC when responding to a request. A SIP message is composed of three parts. The first part of a SIP message is a “request line,” which includes fields to indicate a message method (e.g., INVITE) and a Request URI that identifies the user or service to which the request is being directed. The second part of a SIP message comprises headers whose values are represented as name-value pairs. The third part of a SIP message is the message's body, which is used to describe the session to be initiated or contain data that relates to the session. Message bodies may appear in requests or responses.
User agents can communicate by sending SIP messages during a SIP dialog. A SIP dialog is a peer-to-peer relationship between two user agents that persists for some time. A dialog may be established when a UAC sends an INVITE request to a UAS and the UAS replies with a 200 OK response.
Applications may employ SIP with a lower-level protocol to send or receive messages. SIP may use lower-level connections to transport a dialog's messages, such as Transmission Control Protocol/Internet Protocol (“TCP/IP”), which are commonly employed transport- and network-layer protocols. Transmission Control Protocol (“TCP”) is a connection-oriented, reliable delivery transport layer protocol. TCP is typically described as a transport layer that provides an interface between an application layer (e.g., an application using SIP) and a network layer. The application layer generally communicates with the TCP layer by sending or receiving a stream of data (e.g., a number of bytes of data). TCP organizes this data stream into segments that can be carried by the protocol employed at the network layer, e.g., the Internet Protocol (“IP”). These segments of data are commonly referred to as “packets,” “frames,” or “messages.” Each message generally comprises a header and payload. The header comprises data necessary for routing and interpreting the message. The payload comprises the actual data that is being sent or received. The application, transport, and network layers, together with other layers, are jointly referred to as a data communications stack.
When an initiating participant wants to start a real-time conversation, that participant needs to know whether the intended participants are available to respond in real time to a message. If not, then communications via conventional electronic mail, voice mail, or some other mechanism may be more appropriate. For example, if the computers of the intended participants are currently powered off, then a real-time conversation may not be possible. Moreover, if their computers are currently powered on, but the intended participants are away from their computers, a real-time conversation is also not possible. The initiating participant would like to know the availability of the intended participants so that an appropriate decision on the form of communication can be made.
The availability status of an entity such as a computer system (i.e., endpoint) or a user associated with that computer system is referred to as “presence information.” Presence information identifies the current “presence state” of the user. Users make their presence information available so that other users can decide how best to communicate with them. For example, the presence information may indicate whether a user is logged on (“online”) with an instant messaging server or is logged off (“offline”). Presence information may also provide more detailed information about the availability of the user. For example, even though a user is online, that user may be away from their computer in a meeting. In such a case, the presence state may indicate “online” and “in a meeting.”
In an instant messaging context, a publishing user (“publisher”) may provide their presence information to a presence server that then provides the presence information to subscribing users (“subscribers”). Thus, a presence server may use a subscriber/publisher model to provide the presence information for the users of the presence service. Whenever the presence information of a user changes, the presence server is notified of the change by that user's computer system and in turn notifies the subscribing users of the change. A subscribing user can then decide whether to initiate an instant messaging conversation based on the presence information of the intended participants. For example, if the presence information indicates that a publishing user is currently in a conference telephone call, then the subscribing user may decide to send an instant message, rather than place a telephone call, to the publishing user. If the subscribing user, however, needs to call and speak with the publishing user, the subscribing user needs to monitor the presence information of the publishing user to know when the call can be placed. When the subscribing user notices that the publishing user's presence information indicates that the telephone conference has been concluded, the subscribing user can then place the telephone call. A specification relating to presence information in instant messaging systems, “RFC 2778,” is available via the IETF web site. A draft of a proposed specification relating to presence information in SIP is available at via the IETF web site.
Because of the popularity of instant messaging and other real-time services, a provider of such services often needs to provide a pool of servers to support the real-time service. For example, a large organization may need to provide presence information for tens of thousands of users. Because of the real-time requirements and the high volume of communications that are sent between the clients and servers of a presence service, a single server is typically not sufficient. Although the presence service may be provided by a pool of presence servers, the pool appears as a single presence server to the client. For example, when a client logs on to or establishes a connection with a presence service, the logon request may be routed through a load balancer that selects an available server of the server pool and directs the logon request to the selected server. When a client publishes presence information of its user, the client's server provides the presence information to the client's subscribers.
Connections between clients and servers can be broken for various reasons. For example, if a client is connected to a server via a wireless network, then a problem with the wireless communications may cause a connection to be broken. When a server notices that its connection to a client has been broken, it will typically log off the client (e.g., terminate the application-level connection of that client). When the client notices the broken connection, the application layer of the client is typically notified of the broken connection by a service layer (or stack layer) of the client. The application layer may notify the user of the broken connection and attempt to log back on to the server. When a user with a wireless computing device moves about, the connection with the network can be broken because of gaps in the area serviced by the wireless network, communications errors, interference, and so on. Such users may find it annoying to be notified of every broken connection especially when the user will typically want to immediately re-log on to the server.
A broken connection may also result from a hardware or software failure at the server. Clients that are logged on through a server will typically notice that the connection (e.g., TCP connection) with that server has been broken. Upon detecting the broken connection, each client may try to re-log on to a server. If hundreds or thousands of clients are impacted by the failure and immediately send logon requests to the server pool, the server pool may become overloaded and be unable to service the requests in a timely fashion. Moreover, a client that sends a logon request may detect that the server pool has not responded in a timely manner and re-send the logon request. As a result, the server pool will become even more overloaded in a cascading manner.